Repository navigation
Description
Problem
When a proposal answers a brief with creators or shows, the buyer (a person or an agent) wants a card for each one: a picture, a few sample videos or a sizzle reel, audience stats (for example 55% male, mostly 18–26, 1.2M subscribers, 400K average views) and topics. Most of this has nowhere to go in AdCP today.
| Need | Today |
|---|---|
| Picture or thumbnail | No image field on collection. product_card images are product-scoped. |
| Sample content, sizzle reel | product_card_detailed.reference_assets[] has no sample-content role. Sellers have to use other. |
| Audience composition | audience_evidence defines only an age dimension. Gender needs a provider-specific URI. |
| Subscriber and follower counts, average views | None outside forecast metrics. |
| Topics | collection.genre[] with genre_taxonomy, and installment.topics[]. These are fine. |
There's also no typed link between the channel and the person. talent[].brand_url is a loose URL. The person's likeness and endorsement rights live in brand.json as a rights-holder brand (right_type: talent), but a collection can't point at that brand with a {domain, brand_id} reference. So a buyer can't connect "this channel" (what's being sold) to "this person" (whose rights are being licensed).
Proposal
collection.images[]: thumbnail and banner.collection.sample_content[]:{url, kind, published_at}, plus asample_contentrole forreference_assets.collection.audience_stats: follower count per distribution, average views per installment,measured_atandsource.- A standard
genderdimension inaudience-characteristic. talent[].brand_ref({domain, brand_id}) alongsidebrand_url.
This is the display and audience slice of #1839, which triage asked to split into separate identity, trust and audience issues. Related: #6538.
Metadata
Metadata
Assignees
Labels
Type
Projects
- StatusShow more project fieldsShipped
Activity
bokelley commentedon Oct 9, 2026
Triage
Classification: Feature request (RFC)
Bucket(s): spec / protocol, brand, schema
Status: ready-for-human
Milestone: Spec Backlog
What the experts said:
brand_refcleanly reuses/schemas/core/brand-ref.json;genderneeds a value-set definition and GDPR Art. 9 scoping note before conformance tests can be written;audience_statsneeds an immutability-model decision and a MUST NOT targeting-predicate clause.images[],brand_ref,collection.sample_content[]are clean with zero adoption cost;audience_statsas a standalone object creates fragmentation alongsideaudience-evidence.json— recommends a lightweight seller-declared evidence profile instead;sample_contentrole onreference_assetsis the wrong layer (creative-brief construct, not a seller catalog field — drop it, rely oncollection.sample_content[]alone);genderneeds explicit WG governance sign-off due to GDPR/CCPA special-category implications.My take: Three of five proposals (
images[],collection.sample_content[],talent[].brand_ref) are schema-coherent, non-breaking, and agreed upon — ready for WG ratification as a narrow first PR. Two proposals (audience_statsshape,gendervocabulary) have open design questions that need your call before they can be conformance-tested.Decision 1 —
audience_statsshape — experts disagree; @bokelley, your call: A or B?Option A — standalone lightweight field on
collection(what the RFC proposes):{ "audience_stats": { "distributions": [ { "platform": "youtube", "followers": 1200000 }, { "platform": "instagram", "followers": 430000 } ], "avg_views_per_installment": 400000, "measured_at": "2026-09-01T00:00:00Z", "source": "seller_declared" } }Simple to produce; intentionally mutable/display-only (no hash/attestation ceremony). WG must add a MUST NOT targeting-predicate clause. Introduces a third audience-data bucket alongside
audience-evidenceandext.Option B — seller-declared profile inside existing
audience-evidence[]:{ "evidence_id": "yt-followers-2026q3", "snapshot_id": "snap-001", "content_digest": "sha256:...", "evidence_type": "seller_declared", "relationship": "reach_estimate", "value": 1200000, "unit": "count", "audience": { "dimension": "age", "range": { "min": 0, "max": 99 } }, "measurement_window": { "start": "2026-09-01", "end": "2026-09-30" } }Reuses existing schema machinery; buyer agents see one coherent audience-data shape. Some
audience-evidencerequired fields (baseline,methodology,subject_type,provider) would need to be relaxed for this profile to avoid ceremony overload.Decision 2 —
genderdimension vocabulary — @bokelley, your call: A or B?Option A — canonical AdCP
genderdimension (what the RFC proposes):{ "dimension": "gender", "value": ["male", "female", "nonbinary"] }Simpler wire; maximally interoperable. Requires AdCP to define and own a demographic vocabulary — a protocol stance with GDPR Art. 9 special-category implications that should be an explicit WG ballot, not an incidental schema addition.
Option B — delegate to provider taxonomy (already works today):
{ "dimension": "https://iabtechlab.com/taxonomy/2.1#50001", "value": "male", "taxonomy": { "system": "https://iabtechlab.com/taxonomy/2.1", "version": "2.1" } }AdCP stays neutral on demographic vocabulary; providers choose their taxonomy URI. Less interoperable — buyer agents must know each provider's taxonomy to compare gender splits across sellers.
Recommendation: A narrow first PR for
images[]+brand_ref+collection.sample_content[]can open immediately (all three agreed, clean delta:core/collection.json,core/talent.json; also adddeprecated: truetotalent[].brand_urlper theformat_idsmigration pattern). File a follow-up issue scoped toaudience_stats+genderonce calls 1 and 2 land.This is the display and audience slice of #1839 (Spec Backlog), related to #6538.
Triaged by Claude Code. Session: https://claude.ai/code/session_01QWfeEzJBD435FLZ7jVVvtz
Generated by Claude Code
fix(schema): address review on collection cards wording and tests
bokelley commentedon Oct 9, 2026
WG decision brief (3.3)
Track: consent round. Objections by Fri 2026-10-16 23:59 UTC. Not on the live agenda unless someone objects. Draft PR: #8098.
Recommended shape. No
audience_statsobject. Each card fact uses a surface that already exists:distribution[].followersoncore/collection-distribution.json. It$refsaudience-size(range,precision,as_of) under feature idmedia_buy.collection_cards. It counts accounts, not people, so it is not summed or read as reach.audience_evidenceasseller_declaredandobserved, with the platform asproviderand a platform-audiencebaseline. No schema change. A worked example is in the draft PR.views. No schema change.followersonly and question 2 moves to 3.4.Adopter. Availnet (a creator marketplace) is the single adopter behind the experimental
media_buy.collection_cardssurface.Questions
audience_statsobject withdistribution[].followers(audience-size), composition throughaudience_evidence(seller_declared), and average views through the forecast? Or shipaudience_statsas proposed? Recommend yes. One audience-data shape per meaning, reusing feat(schema): report audience size as a range on sync_audiences #8024's range-and-precision model.genderdimension toaudience-characteristicusing feat(schema): add experimental portable core gender targeting #7925's value set (female,male,non_binary, hoisted toenums/gender.json)? Or keep provider URIs only? Recommend yes. Population-level composition is not personal data. feat(schema): add experimental portable core gender targeting #7925'ssex_genderrestricted-attribute classification governs targeting, not evidence. Comparing gender splits across sellers is the point of a card. Evidence stays non-targeting, and feat(schema): add experimental portable core gender targeting #7925'sby_demographicstays age-only.viewsmetric an acceptable home for "average views per installment" in 3.3? Or does the WG want a trailing-window metric (median views over the last N installments, with the window stated)? Recommend yes for 3.3. It is a projection for the proposed buy, not a trailing average, and the forecast has no per-installment unit. The draft says so and tells sellers to state the basis in the product description. Revisit in 3.4 if adopters need a measured trailing figure.Open points noted in the PR:
followerslives on collection objects, which publishers author inadagents.json. That has no capabilities response, so theexperimental_featuresrule is worded for a seller that serves or relays collections.audience_evidencecarries digest-pinned snapshots, which is heavy per creator.Default if no objection by Fri 2026-10-16 23:59 UTC: the recommendation as drafted.
docs: name Availnet as the collection cards adopter
docs: name Availnet as the adopter for collection cards
bokelley commentedon Oct 11, 2026
Slice C filed as #8137: return full collection objects in get_products and proposal responses, add
collection.url, and resolve people throughtalent[].brand_refto their brand or rights agent (get_brand_identity) rather than copying identity into the collection.feat(schema): experimental collection cards with images, sample conte…
feat(schema): experimental collection objects in product and proposal…
feat(schema): experimental follower count on collection distribution …
bokelley commentedon Nov 6, 2026
Done for 3.3, as experimental: collection cards in #8094, follower count in #8098, and full collection objects in #8141 (slice C, #8137). Closing.